COMPENSATING WAGE DIFFERENTIALS FOR RISK OF
DEATH IN GREAT BRITAIN: AN EXAMINATION OF THE
TRADE UNION AND HEALTH AND SAFETY COMMITTEE
IMPACT
S. Grazier
University of Wales Swansea
ABSTRACT
British estimates of the compensating wage differential for injury risk have centred
around 1980s data; the nature and prevalence of accidents in the workplace have
however, changed substantially since then. This paper, using recent British data, finds
a significant wage premium for exposure to risk of fatality at work. Research in this
area has pointed to the ambiguous effect that trade unions have upon the risk
premium, and a re-examination of the union impact points to a negative effect. With
an emerging literature investigating the relationship between other forms of worker
representation and injuries, consideration is also given to the arrangements for dealing
with health and safety at the workplace level. Health and safety committees
established solely to discuss health and safety issues are found to have an
independent, positive effect upon the risk premium. Evidence therefore supports the
notion that the role of health and safety committees should be considered separately
from any trade union effect.
JEL Classification: J0, J3, J5
Keywords: Compensating wage differentials, Accident risk, Trade unions, Health and
safety committees
Contact Details: Suzanne Grazier, School of Business and Economics, Swansea
University, Richard Price Building, Singleton Park, Swansea, SA2 8PP
E mail: 166261@swansea.ac.uk
1
1.
INTRODUCTION
The theory of compensating wage differentials has been investigated by many papers,
especially with regard to whether workers receive a wage premium for exposure to
high accident risk 1 . In such models, the compensating wage differential would be the
price employers are required to pay workers in order for them to accept employment
with an increased risk of fatality or injury. Such findings have a direct policy
application; estimates of a risk premium can be used to calculate the Value of a
Statistical Life (VSL) or Injury (VSI) which can be applied to evaluate many public
policies in areas such as the environment and road safety 2 .
Marin and Psacharopoulos (1982), in the first paper using British data from the Office
of Population Censuses and Surveys (OPCS) Occupational Mortality Decennial
Supplement 1970-72, find evidence of a wage premium for exposure to fatal risk.
Sandy and Elliott (1996) and Arabsheibani and Marin (2000) using similar data over
the period 1979 to 1983, and Siebert and Wei (1994) using Health and Safety
Executive (HSE) data for 1986 to 1988, all find evidence of a fatal risk premium.
There are no papers however, that attempt to estimate the premium using more recent
British data.
The existence of a wage premium for exposure to risk of non-fatal injury is uncertain,
with no UK study finding evidence that workers are compensated for this.
Arabsheibani and Marin (2000) include a non-fatal injury variable in their estimation
based on a sample of male manual workers using data from the General Household
Survey (GHS), which asks respondents if they have had an accident at work that
1
Recently, Sandy and Elliott (2005) estimate a significant compensating differential for exposure to
high illness risk.
2
See Ashenfelter (2006) for a full discussion.
2
resulted in a visit to a doctor or hospital. No evidence of a premium is found, which
they attribute to the fact that their variable does not distinguish the degree of severity
of an accident. Siebert and Wei (1994) again for a sample of male manual workers,
include a non-fatal injury variable defined as an accident that resulted in absence from
work for 3 or more days using HSE data. This also turns out to be insignificant.
There has been particular interest in the effect that trade unions have upon the risk
premium. Theoretically trade unions could increase the risk premium, through
improved information collection and collective bargaining, or reduce it, if unions are
more concerned with reducing risk rather than increasing compensation for exposure
to it. US studies tend to find that unions increase the premium while British studies
tend to find they reduce it, Siebert and Wei being the exception. Given the decline in
union membership since the 1980s 3 , when most studies analysed the issue, the impact
trade unions have upon the risk premium seems to require further investigation.
In recent years, investigations into the effect unions have upon safety in the workplace
have also tended to consider firms arrangements for dealing with health and safety
within the workplace 4 . Under current legislation firms are not obliged to appoint
safety representatives or establish health and safety committees. However, safety
representatives can now be appointed in firms with no union presence under the
Health and Safety (Consultation with Employees) Regulations 1996. If two or more
safety representatives request a committee be established, a firm must do so within
three months. Health and Safety committees are mandatory for firms over a certain
3
4
See Machin (2000)
See Fenn and Ashby (2004) p.464
3
size in both France and Germany, and as Reilly et al. (1995) note EU legislation may
eventually require this elsewhere.
Reilly et al. (1995) suggest that in Britain health and safety committees may adopt an
important role “given the potential for a continued decline in union workplace
strength” (p.276). Supporting this, they find that firms with such committees have on
average fewer injuries. They claim establishments with joint consultative committees
exclusively for health and safety and with all employee representatives chosen by
unions, have on average 5.7 fewer injuries per 1000 employees compared to
establishments where management deals with health and safety without consultation
(p.283). This paper has been extremely influential, with their findings widely cited in
support of the beneficial effects that trade unions and health and safety committees
have upon workplace safety 5 .
In a replication of Reilly et al. (1995) however, the evidence provided by Nichols et
al. (2005) is less certain. Although their results support the view that health and safety
should not be left to management alone, they find no evidence to support their more
precise conclusions (p.25). Related to this, Fenn and Ashby (2004) find conflicting
evidence that the presence of health and safety committees is associated with a higher
number of injuries, which they attribute to improved and greater awareness of
accident reporting procedures. Nichols et al. go on to conclude that “there is good
cause to re-examine a whole number of issues and dynamics that may affect the
determination of health and safety” (p.26). Litwin (2000), in a study of the effect trade
unions have upon industrial injury, emphasises the need to “separate the effect of
5
Nichols et al. (2005) present extracts from numerous policy documents in which the findings of Reilly
et al. are cited.
4
health and safety committees and joint consultative committees from the effects of the
variables of workplace union strength” (p.5). The effect, if any, that health and safety
committees may have upon the accident risk premium has not yet however, been
investigated. Given the growth of such institutions in the British labour market, such
analysis is timely.
Compensating wage differential research has been constrained by problems of
endogeneity and unobserved heterogeneity. Risk may be endogenously determined
with wages; if safety is a normal good, we would expect those with greater earnings
potential to choose safer jobs. In addition, unobserved heterogeneity may influence
the premium for job risk if some individuals possess unobserved qualities that affect
their ability to work in risky jobs. There is disagreement over the effect this will have
upon estimates. While Hwang, Reed and Hubbard (1992) find such measurement
error leads to a downward bias in the risk premium, Shogren and Stamland’s (2002)
analysis suggests the risk premium will be overestimated. Although Garen (1988)
formulates a model to control for this bias, his method involves finding instrumental
variables that proxy risk aversion, and this has proved problematic. Weak instruments
are found to lead to biased estimates close to the original OLS estimates by Bound et
al. (1995). Consequently, Bell et al. (2004) describe controlling for unobserved
heterogeneity as “the greatest challenge facing researchers in estimating
compensating wage differentials for workplace risks” (p.1).
This paper uses recent data to estimate whether workers receive a compensating wage
differential for exposure to fatal accident risk in Britain. Using data which
distinguishes major injuries from less severe accidents, we also consider whether a
5
risk premium is found for non-fatal accident risk. In addition, particular attention is
given to the impact that trade unions and workplace health and safety committees may
have upon the risk premium.
2.
METHODOLOGY
A standard wage equation is estimated [1], where Yi denotes the earnings of the ith
individual, X is a vector of other determinants of earnings, Di is a measure of fatal
and/or non-fatal risk in individual i’s occupation, the interaction term Ui Di denotes
the impact of unions on the risk premium, and εi is a random error term which has an
expected value of zero with zero covariance.
Ln Yi = β0 + β1 Xi + β2 Di + β3 Di2 + β4 Ui Di + εi
[1]
A positive and significant β2 coefficient indicates a premium is received for exposure
to risk. We are also interested in whether the wage-risk trade-off takes a linear,
convex or concave form, and to test this, Di² is often included in equation 1.
If positive and significant risk coefficients are estimated, they can be used to calculate
VSL and VSI estimates. Such measures have been criticised 6 on the grounds that
estimates are so wide-ranging that this reduces their usefulness as a guide for policy.
Equation 2 depicts the VSL / I formula, assuming risk is measured per 1000 workers.
VSL / I = (Average Annual Income) (Risk Paramerter * 1000)
6
[2]
Viscusi and Aldy (2003) discuss these criticisms in detail.
6
Viscusi and Aldy (2003) report estimates from a range of studies in terms of US
dollars for the year 2000, and so to enable comparison with the literature, VSL/I
estimates are also reported in this way 7 .
Workers may select into firms covered by trade unions and hence union status may
be non-random. The Heckman Selectivity Correction (1979) is employed to control
for sample selectivity. A probit equation for union status is estimated with a vector of
instruments that determine union status but which are uncorrelated with earnings.
These results are used to calculate the Inverse Mills Ratio [3]:
λ (t) = - f (t) / F (t)
[3]
f is the standard normal density function, F the cumulated normal and t = γ ’ y
calculated from the probit with y the explanatory variables. Lambda is included as an
explanatory variable in the wage equation, with a significant lambda coefficient
indicating union selection is a problem in the estimation. As there is also the potential
for the presence of a health and safety committee to be endogenous within a firm,
with employees in risky occupations choosing to join a firm where there is one, the
same method is employed for selection into health and safety committees.
The previous literature has highlighted endogeneity of risk as being one of the
greatest problems for research in this area. There are two problems: individuals with
higher earnings potential are likely to choose safer jobs, and there is unobserved
7
Estimates are converted using Officer and Williamson (2006)
7
heterogeneity that affects productivity and therefore earnings in risky jobs.
Consequently, there is a cross-equation correlation of disturbances in the wage and
risk equations. Heckman’s method cannot be used here because risk is a continuous
variable. Garen (1988) proposes an instrumental variables method for obtaining
unbiased estimates of the compensating wage differential and this has been
extensively used in the subsequent literature. Considering only fatal risk, the first
stage involves estimating a risk equation [4], where Zi proxies risk aversion, Li is nonwage income, and μ is unobserved heterogeneity.
D i = δ 0 + δ1 Xi + δ2 Zi + δ3 Li + μi
[4]
The disturbance term μ may depend on the wage equation [1] disturbances as workers
with unobservable characteristics that make them more productive in risky jobs will
choose higher D. The second stage involves estimating equation 5, which uses the
disturbances obtained through the risk estimation:
Ln Yi = β0 + β1 Xi + β2 Di + γ1 μi’ + γ2 μi’ Di + θi
[5]
Garen shows estimating equation 5 will yield consistent estimates.
3.
DATA
Data from the HSE are used to measure accident risk and the Labour Force Survey
(LFS) to estimate numbers of workers in each occupation. The Workplace
Employment Relations Survey 2004 (WERS 04) is used for matched data on workers
and workplace characteristics.
8
The Reporting of Injuries, Diseases and Dangerous Occurrences Regulations 1995
(RIDDOR 95) places a legal requirement upon employers in Britain to report specific
incidences of fatalities and injuries at work to the HSE or local authority. Statistics
available from the HSE on work fatalities, major injuries and over 3-day injuries are
compiled from reports made under this regulation. RIDDOR 95 states employers must
report incidences of an accident resulting in death or major injury 8 arising out of, or in
connection with, work. Employers are also required to report accidents that result in
an employee being incapacitated from work for more than 3 consecutive days 9 .
Fatal risk is calculated across occupations following Sandy et al. (2001) who find this
is superior to assigning risk by industry or by a mix of industry and occupation codes.
HSE data for 2002/03, 2003/04 and 2004/05 are utilised. The number of accidents
over this three year period is used given that fatal accidents are rare events. A fatal
risk variable is calculated for each occupation as a rate per 1000 workers, as shown by
equation 6:
⎡ Fatalities at Work 2002 / 03 − 2004 / 05 ⎤
Fatal Risk = ⎢
⎥ * 1000
⎣ Number employed in occupation ⎦
[6]
The LFS is used to provide data on the number of workers employed in each
occupation over the same period as the HSE data. Risk is calculated for each 3 digit
Standard Occupational Classification 2000 (SOC 2000) giving a total of 81
occupations. In addition, a variable is constructed for Major Injury Risk.
8
See HSE (2006) for the full list of injuries that are reportable under the Major Injuries category.
Incidences that are not reportable include road traffic accidents that involve people travelling in the
course of work which is covered by road traffic legislation. Accidents to members of the Armed Forces
and injuries to the self-employed due to an accident at their own premises are also excluded.
9
9
WERS 2004 is used to provide data on employees. As a matched employer-employee
survey, WERS provides detailed information on employee personal characteristics,
the nature of their work and their attitudes towards their job, but also provides
manager reported workplace data. Risk variables are assigned to workers in WERS
via occupation codes.
Estimations are usually carried out for male manual workers, because an accident
premium is most likely to be found in samples of workers who are exposed to the
greatest risk. The sample therefore, is divided into manual and non-manual
occupations 10 , leaving 33 manual occupations. Regressions are estimated for two
samples: all manual workers and male manual workers 11 . Estimations are usually
restricted to men because of problems of measuring risk for women, who are less
likely to be found in more risky occupations. However, the risk data used are
applicable to both men and women and so a sample that includes women is tested.
Those that work less than 30 hours per week are excluded as these workers may be
exposed to less risk than that captured by the risk variable. Table 1 presents means
and standard deviations for the three risk variables in each sample.
Siebert and Wei (1994) calculate a mean fatal risk of 0.038 per 1000 workers for their
sample of male manual workers, and Sandy and Elliott (1996) calculate a mean fatal
risk of 0.044 per 1000 male manual workers. Derived fatal accident rates do not differ
considerably from the earlier literature therefore. In terms of non-fatal risk, Siebert
10
The following occupations are classed as manual: 51 skilled agricultural trades, 52 skilled metal and
electrical trades, 53 skilled construction and building trades, 54 textiles, printing and other skilled
trades, 61 caring personal service occupations, 62 leisure and other personal service occupations, 81
process, plant and machine operatives, 82 transport and mobile machine drivers and operatives, 91
elementary trades, plant and storage related occupations, 92 elementary service occupations.
11
Estimations were also conducted for non-manual workers with no significant risk premiums found,
as predicted.
10
and Wei derive a variable that encompasses major and over 3 day injuries with a
mean value of 14.246 per 1000 workers for their male manual sample. For the
equivalent sample, a mean major injury risk of 5.65 per 1000 is calculated. Hence,
major injury risk is found to be much lower than their variable which also
encompassed less serious injuries, emphasising the difference between the two.
Table 1: Risk Variable Descriptive Statistics (per 1000 workers)
ALL MANUAL
Number
Mean
Standard Deviation
MALE MANUAL
Number
Mean
Standard Deviation
Fatal
Major Injury
5580
0.0328
0.0382
5580
5.0212
3.8317
3956
0.0412
0.0399
3956
5.6499
3.8043
The fatal risk rates include 6 out of the 33 manual occupations that have been
assigned a zero value. Sandy et al. (2001) suggest assigning the average value of this
variable to such occupations. They believe it unlikely these occupations have no
accident risk, and suggest the zero rate occurs because of data problems, with no
accidents occurring over the time period taken. A 3 year period however, is fairly
extensive, and considered to result in a fairly accurate picture of the degree of
riskiness of occupations. Sandy et al. also acknowledge that assigning average risk to
such occupations makes little difference, as this value is very close to zero. Here,
assigning average fatal risk to zero fatal risk occupations increases the mean fatal risk
rate from 0.0412 per 1000 workers to 0.0422 per 1000 workers for the sample of male
manual workers. Although estimations will be carried out without assigning average
risk to zero risk occupations, this will be considered for comparison purposes.
11
The dependent variable is based on WERS, which asks employees to consider their
average weekly pay before tax and other deductions. Each worker has a choice of 14
possible pay brackets, so interval regression is used for the estimation. To calculate
VSL and VSI we need to know the average annual income for each sample. The same
question used to formulate the dependent variables is used, only the mid point of the
mean wage bracket is taken as the average income, and multiplied by 52 to give a
yearly figure. The resulting variables are Wkincome and Anincome
WERS is used to construct explanatory variables similar to those used in the earlier
literature. Appendix 1 defines the variables, which are taken from both the employee
and management surveys, and reports descriptive statistics.
Two trade union variables are constructed. Unioncov is derived from the employee
survey and indicates whether the workplace has a union presence. Appendix 1 reports
that for the sample of all manual workers, 52.6 per cent of workplaces have a trade
union presence, compared to 56.1 per cent in the male manual sample. A further trade
union variable, Runion, is a dummy drawn from the management survey according to
whether managers have reported recognising a union. Descriptive statistics reveal
differences between the two variables, with Runion having a greater mean in both
samples, suggesting some workers do not realise their workplace has a union
presence. We would expect Runion to give the most accurate reflection.
Further variables have been constructed to denote arrangements for dealing with
health and safety in the workplace. Reilly et al. (1995) construct 8 variables with
some having very small means. One of the criticisms made by Nichols et al. of the
Reilly et al. study is that there are too many variables covering the organisation of the
12
arrangement of health and safety committees. Therefore, we construct four variables
by merging the Reilly et al. variables 12 . Of these, the main variables tested in the
estimation are Commspecific, which denotes a workplace that has a committee
exclusively for health and safety, and Commgen which denotes a workplace that has a
committee that deals with a range of issues in addition to health and safety. Whilst
41.9 per cent of manual workplaces in the sample have a committee that deals
specifically with health and safety issues, 27.4 per cent do not consult with employees
regarding health and safety matters.
4.
ESTIMATION
Interval regressions are first estimated with just the fatal risk variable, as many studies
in the literature do not include a non-fatal injury variable. Table 2 presents results.
Working overtime, being a supervisor, being a permanent employee and having
worked for a firm for more than a year are all associated with a greater wage.
Working for a large firm is significantly associated with greater pay, with nemps²
negative and significant indicating a concave relationship. Runion and Commspecific
are positive and significant suggesting unions and health and safety committees have
a positive impact upon wages overall 13 .
12
In terms of the Reilly et al variables:
Commspecific=Hs1+Hs2+Hs3; Commgeneral=Hs4+Hs5+Hs6; Emprep=Hs7;Nohsconsult=Hs8
13
The presence of health and safety committees may indicate a firms’ commitment to consultation,
both in terms workplace safety but also in terms of wages.
13
Table 2: Interval Regression Estimates (Fatal risk variable only)
Dependent variables: Lnwpayl Lnwpayh
MANUALWORKERS
Constant
4.8388***
(0.0443)
Educ1
0.1270***
(0.0105)
Educ2
0.0466***
(0.0160)
Educ3
0.2123***
(0.0157)
Tenure2
-0.0002
(0.0190)
Tenure3
0.0484***
(0.0165)
Tenure4
0.0686***
(0.0165)
Tenure5
0.1430***
(0.0164)
Overtime
0.0066***
(0.0008)
Flexitime
-0.0008
(0.0109)
Supervise
0.1336***
(0.0119)
Runion
0.0904***
(0.0144)
Commspecific
0.0408***
(0.0103)
Permanent
0.0668***
(0.0230)
Age
0.2234***
(0.0171)
Age2
-0.0212***
(0.0019)
Nemps
4.64e-05***
(1.33e-05)
Nemps2
-5.91e-09**
(2.51e-09)
Meritpay
0.0545***
(0.0114)
Public
-0.0737***
(0.0123)
Female
-0.3285***
(0.0115)
Fatal
0.6456***
(0.1828)
Fatal*Runion
-0.3789**
(0.3643)
Obs
Wald chi2
Log pseudo likelihood
VSL(2004 £)
VSL (2000 US$)
5474
3092.41
-10988.213
£10,557,277
$15,194,431
MALE MANUAL WORKERS
4.7005***
(0.0584)
0.1224***
(0.0122)
0.0400**
(0.0188)
0.2090***
(0.0215)
-0.0210
(0.0225)
0.0531***
(0.0183)
0.0608***
(0.0191)
0.1422***
(0.0189)
0.0076***
(0.0010)
-0.0018
(0.0129)
0.1255***
(0.0138)
0.0600***
(0.0177)
0.0477***
(0.0119)
0.0696**
(0.0285)
0.2815***
(0.0223)
-0.0259***
(0.0024)
5.47e-05***
(1.58e-05)
-7.30e-09**
(3.25e-09)
0.0481***
(0.0129)
-0.1136***
(0.0145)
0.5074**
(0.2028)
-0.1345
(0.2893)
3897
1263.89
-7784.232
£9,163,664
$13,188,691
14
Results indicate there is a wage premium for being exposed to risk of death in the
workplace, with a positive coefficient estimated for Fatal which is significant at the 1
per cent level in both samples. The all manual workers sample produces the greatest
wage premium, with a VSL of £10.6 million compared to £9.2 million for the male
manual sample. The premium is likely to be larger in the all manual sample because
of the inclusion of women, who are exposed to less risk and are more averse to risk,
and hence require a larger wage premium per unit of risk14 . The union-risk interaction
variable Fatal*Runion is negative but only significantly so in the all manual sample.
This suggests trade unions have the effect of reducing the risk of death premium, as
found in most of the UK literature. The estimation is repeated including a Fatal²
variable: the Fatal² coefficient is significantly negative in both samples indicating the
relationship between wages and fatal risk is concave. This is consistent with findings
in the earlier literature.
The estimation is repeated including in addition to Fatal, the Major Injury variable.
Key results are reported in Table 3.
A premium for exposure to risk of death still remains when Major Injury is included
in the manual workers and male manual workers sample. Major Injury however, is
significantly negative in both samples, suggesting no premium is received for
exposure to non-fatal injury risk. This finding is similar to that found in other
compensating wage differential studies using UK data.
14
Deleire and Levy (2004) discuss this issue, and find evidence that the risk of death associated with an
occupation has a greater negative effect upon occupation choice for women compared to men.
15
Table 3: Interval Regression Results (Fatal and Major Injury)
Dependent Variables: Lnwpayl Lnwpayh
MANUAL WORKERS
Constant
4.8544***
(0.0443)
Runion
0.0921***
(0.0144)
Commspecific
0.0447***
(0.0103)
Female
-0.3298***
(0.0115)
Fatal
1.0374***
(0.2160)
Major Injury
-0.0066***
(0.0016)
Fatal*Runion
-0.3848
(0.2620)
Obs
Wald chi2
Log
pseudo
likelihood
VSL (2004 £)
VSL (2000 US$)
MALE MANUAL WORKERS
4.7244***
(0.0583)
0.0622***
(0.0179)
0.0523***
(0.0119)
0.9343***
(0.2309)
-0.0085***
(0.0018)
-0.1191
(0.3006)
5474
3119.68
-10978.661
3897
1301.19
-7770.938
£16,964,249
$24,415,587
£16,873,495
$24,284,970
Other variables included in estimation: Educ1, Educ2, Educ3, Tenure2 Tenure3, Tenure4, Tenure5,
Overtime, Flexitime, Supervise, Security, Permanent, Age, Age², Nemps, Nemps², Meritpay, Public
Interval regressions are also performed assigning average risk to the variable Fatal.
Fatal remains positive and significant in both samples, with the magnitude of the
variable slightly smaller when average values are assigned.
TRADE UNIONS
The role that trade unions have upon the risk premium is examined further. Results
have indicated trade unions have a negative effect upon the risk premium. This
conclusion remains when the variable Unioncov, taken from the employee
questionnaire, is included in the estimations in replace of Runion.
16
An alternative way to consider the effect that trade unions have upon the premium for
injury risk is to split the sample of workers according to their union coverage status.
Although the usual method in the literature is to use a risk*union interaction variable,
Siebert and Wei use this sample splitting method, and Fairris (1992) argues that
estimating separate equations by union status is the most appropriate method because
of “important institutional differences between the union and non-union sectors”
(p.266).
Table 4: Descriptive Statistics by Union Status (Mean and Standard Deviation)
Wkincome
Anincome
Wpayl
Wpayh
Commspecific
Fatal
Major
ALL MANUAL
Covered
Uncovered
333.1531
290.8539
(132.472)
(137.1949)
17323.96
15124.40
(6888.543)
(7134.136)
302.7724
263.6879
(118.2643)
(125.1147)
361.7016
314.4329
(142.9971)
(143.2844)
0.5558
0.2463
(0.4970)
(0.4309)
0.0310
0.0350
(0.0287)
(0.0391)
5.0110
5.0293
(3.8257)
(3.8399)
MALE MANUAL
Covered
Uncovered
362.3650
326.7213
(129.8704)
(137.4774)
18842.98
16989.51
(6753.262)
(7148.827)
329.1103
296.7959
(115.2189)
(124.56)
393.2336
352.3882
(140.5282)
(143.8491)
0.5929
0.2473
(0.4914)
(0.4316)
0.03829
0.0451
(0.0391)
(0.0407)
5.6344
5.6638
(3.8227)
(3.7803)
The two samples are divided according to the variable Runion; workers who are
employed by a firm where managers have indicated they do recognise a union are
assigned to the covered sector, those that do not are assigned to the uncovered sector.
Table 4 illustrates that workers covered by union terms and conditions receive higher
pay in both samples, which is consistent with the positive Runion estimates in the
wage regressions. Workers in the covered sector are shown to face on average a
slightly lower risk of death. This is consistent with the argument that unions are
17
concerned with increasing safety in the workplace rather than increasing the
compensation for risk.
Table 5: Interval Regression Results (Fatal)
Dependent Variables: Lnwpayl Lnwpayh
ALL MANUAL
Covered
Uncovered
Constant
4.9403***
4.8263***
(0.0626)
(0.0612)
Commspecific
0.0345***
0.0432**
(0.0131)
(0.0174)
Fatal
0.3564*
0.5054***
(0.1931)
(0.1906)
Obs
Wald chi2
Log
pseudo
likelihood
VSL (2004 £)
VSL (2000 US$)
MALE MANUAL
Covered
Uncovered
4.8599***
4.6304***
(0.0758)
(0.0829)
0.0409***
0.0622***
(0.0145)
(0.0214)
0.3489*
0.4910**
(0.2057)
(0.2054)
3058
1518.49
-6065.5404
2416
1342.14
-4890.8649
2251
698.28
-4395.5211
1646
523.08
-3363.5314
£5,828,088
$8,388,004
£8,264,634
$11,894,773
£6,301,148
$9,068,850
£8,867,480
$12,762,411
Other variables included in estimation: Educ1, Educ2, Educ3, Tenure2 Tenure3, Tenure4, Tenure5,
Overtime, Flexitime, Supervise, Permanent, Age, Age², Nemps, Nemps², Meritpay, Public, Female (all
manual sample)
Splitting the sample by union status suggests that trade unions have a negative effect
upon the fatal risk premium (Table 5), as found in other studies that included an
interaction variable (e.g. Marin and Psacharopoulos 1982, Sandy and Elliott 1996).
Greater premiums are found for uncovered workers in both samples. This reinforces
earlier conclusions that trade unions reduce the fatal risk premium.
HEALTH AND SAFETY ARRANGEMENTS
Having found that trade unions are associated with a lower risk premium, we now
estimate the effect that the presence of health and safety committees may have upon
the risk premium. Runion and Commspecific are positively correlated indicating
18
health and safety committees are more likely in unionised firms, but only at around 34
per cent. It is therefore possible to consider separately the effect upon the risk
premium.
Table 6: Interval Regression (Fatal, Commspecific)
Dependent Variables: Lnwpayl Lnwpayh
ALL MANUAL
Constant
4.8427***
(0.0444)
Runion
0.0940***
(0.0145)
Commspecific
0.0227*
(0.0138)
Female
-0.3288***
(0.0115)
Fatal
0.5492***
(0.1921)
Fatal*Runion
-0.5244*
(0.2725)
Fatal*Commspecific
0.5582**
(0.2725)
Obs
Wald chi2
Log pseudo likelihood
VSL (2004 £)
VSL (2000 US$)
5474
3108.04
-10986.093
£8,980,881
$12,925,622
MALE MANUAL
4.7042***
(0.0586)
0.0634***
(0.0181)
0.0327*
(0.0173)
0.4445**
(0.2124)
-0.2298
(0.3098)
0.3876*
(0.2021)
3897
1264.37
-7783.4355
£8,027,688
$11,553,751
Other variables included in estimation: Educ1, Educ2, Educ3, Tenure2 Tenure3, Tenure4, Tenure5,
Overtime, Flexitime, Supervise, Permanent, Age, Age², Nemps, Nemps², Meritpay, Public
Key interval regression estimates (Table 6) indicate that unions reduce the fatal
premium, as found earlier. However, health and safety committees appear to increase
the fatal premium, with the variable Fatal*Commspecific significantly positive.
Interval regressions are also estimated including in addition the Fatal*Commspecific
interaction variable, a Fatal*Commgeneral variable to capture the effect that
committees that deal with a range of issues have upon the risk premium. See
Appendix 2. Whilst Fatal*Commspecific remains significantly positive in the all
manual worker sample, Fatal*Commgeneral is insignificant. In terms of risk
19
compensation therefore, it is only committees that deal specifically with health and
safety that have an independent impact. A significantly positive coefficient estimated
for Commgeneral however, indicates a positive impact upon wages, again supporting
the positive union wage effect 15 .
To consider this further, the samples are divided according to whether a specific
health and safety committee is present in a workplace. Descriptive statistics reveal
that workers employed by a firm that does have a health and safety committee have
slightly lower mean Fatal and Major Injury risk. Through descriptive statistical
analysis only therefore, health and safety committees do appear to be associated with
fewer workplace accidents. This in part can be attributed to the fact that such
committees are more likely in firms that are covered by union terms and conditions
(although the correlation coefficient is small) as it has also been shown accidents are
less likely in such firms. The finding that trade unions and health and safety
committees have the same negative effect upon accident rates, but have a different
effect upon the risk premium however, is particularly interesting. It could be that such
committees have a positive impact upon the compensation bargaining environment. If
committees work to highlight and disseminate information concerning the injury risk
of certain occupations, they could strengthen the case for higher compensation.
Table 7 presents key interval regression estimates dividing workers according to
whether a specific health and safety committee is present in the workplace.
15
Regressions are also estimated with a Fatal*Emprep variable. No significant result is found,
indicating workplaces with safety representative have no independent effect upon the premium.
20
Table 7: Interval Regression Results (split by Commspecific)
Dependent Variables: Lnwpayl Lnwpayh
ALL MANUAL
Commspecific No Commspecific
Constant
4.9269***
4.8277***
(0.0615)
(0.0589)
Runion
0.0631***
0.0957***
(0.0235)
(0.0190)
Female
-0.3306***
-0.3261***
(0.0183)
(0.0147)
Fatal
0.6816***
0.5179
(0.2077)
(0.3703)
Fatal*Runion
-0.7605**
0.2248
(0.3323)
(0.4346)
Obs
Wald chi2
Log
pseudo
likelihood
VSL (2004£)
VSL (2000$)
2299
1349.84
-4532.1895
3175
1595.11
-6433.9152
£11,145,973
$16,041,705
MALE MANUAL
Commspecific No Commspecific
4.8129***
4.6770***
(0.0704)
(0.0814)
0.0300
0.0761***
(0.0294)
(0.0232)
0.5232**
(0.2261)
-0.4417
(0.30)
0.3801
(0.4244)
0.3651
(0.5023)
1745
600.76
-3403.9261
2152
638.77
-4363.2703
£9,449,013
$13,599,376
Other variables included in estimation: Educ1, Educ2, Educ3, Tenure2 Tenure3, Tenure4, Tenure5,
Overtime, Flexitime, Supervise, Permanent, Age, Age², Nemps, Nemps², Meritpay, Public
The above supports earlier conclusions; higher fatal risk premiums are estimated for
workers employed in firms where there is a specific health and safety committee
compared to those where no such committee is present. Given the changing nature of
industrial relations in Britain, the role of ensuring risk compensation in the form of a
wage premium is received may be better investigated by examining the role of health
and safety committees in addition to the union effect.
5.
MEASUREMENT ERROR
HECKMAN SELECTIVITY CORRECTION
The Heckman selection correction is employed to correct for potential endogeneity of
union presence within a firm. Instruments typically reflect management and worker
21
attitude towards trade unions, which are available in WERS. The union probit is
estimated with instruments that are found to be uncorrelated with earnings 16 . Lambda
is significant at the 5 per cent level in the all manual sample suggesting a problem of
union selection, although it is insignificant in the male manual sample. Fatal remains
significantly positive and Fatal*Runion significantly negative when Lambda is
included. Conclusions as to the union effect therefore, remain unchanged.
As with trade unions, the presence of a health and safety committee in a firm may be
endogenous, with workers employed in risky occupations selecting into firms with
such a committee 17 . Instrumental variables that predict whether a respondent works
for a firm that has a specific health and safety committee but do not predict wages are
constructed. Instruments proxy management attitude towards health and safety
consultation and communication. The Heckman selectivity correction term is
insignificant in both samples, indicating there is not a problem with selection here.
RISK ENDOGENEITY
Garen’s method is employed to control for potential endogeneity of risk, as
highlighted in the methodology. A risk equation with explanatory variables including
variables for non-labour income, and proxies for a workers’ degree of risk aversion is
estimated. As Garen acknowledges “finding proxies for the degree of risk aversion is
a difficult task” (p.12). Measures of the stability of an individual’s lifestyle are
frequently used, assuming they are inversely correlated with the degree of aversion to
risk. These include household income other than wages, marital status, house value,
16
For reasons of space, results are not reported but are available.
Specific health and safety committees only are considered as they appear to have the greatest
independent effect upon the risk premium.
17
22
and number of dependents. Some variables that are often used in the risk estimation
are not in WERS, including whether the respondent is a house owner, partners’
schooling, and whether their partner works. Essentially therefore, there is a lack of
variables to give an indication of non-labour income. The WERS management survey
does include however, questions for the largest occupational group on access to an
employer pension scheme, company car or car allowance, and private health
insurance. These non-monetary variables may provide some approximation to
workers’ non-wage wealth. Industry dummies are also included in the risk
estimations, as in the previous literature 18 .
Appendix 3 lists the instrumental variables and descriptive statistics. Instrumental
variables are included in fatal risk regressions, with key results reported in Table 8.
The R² of the fatal injury regressions are 26 per cent and 17 per cent, which although
small, is similar to those usually reported in the literature 19 . In terms of the
instruments, married is negative but insignificant as commonly reported. The
variables children and disability are also insignificant. Pension and health insurance
are both significantly negative and may provide some proxy for risk aversion.
18
Although there may be some argument for including industry dummies in the wage regressions, they
are often excluded, as the risk coefficients can be sensitive to their inclusion.
19
For example Sandy and Elliott (1996) report R² of 17 per cent for their fatal estimation.
23
Table 8: Risk Regression Results
Dependent Varable: Fatal
ALL MANUAL
Constant
0.0332***
(0.0039)
Runion
-0.0008
(0.0012)
Commspecific
0.0012
(0.0011)
Female
-0.0184***
(0.0012)
Married
-0.0002
(0.0010)
Children
0.0023
(0.0378)
Disability
0.0009
(0.0013)
Pension
-0.0039***
(0.0012)
Car
0.0062***
(0.0015)
Healthins
-0.0040***
(0.0014)
Number of obs
5563
F
51.79
R2
0.2575
Adj R2
0.2526
MALE MANUAL
0.0338***
(0.0053)
-0.0007
(0.0015)
0.0011
(0.0014)
-0.0004
(0.0015)
0.0027
(0.0411)
0.0009
(0.0017)
-0.0055***
(0.0017)
0.0081***
(0.0020)
-0.0037**
(0.0018)
3809
22.78
0.1786
0.1707
Other variables included in estimation: Educ1, Educ2, Educ3, Tenure2 Tenure3, Tenure4, Tenure5,
Overtime, Flexitime, Supervise, Permanent, Age, Age², Nemps, Nemps², Meritpay, Public, Ind1, Ind2,
Ind3, Ind5, Ind6, Ind7, Ind8, Ind9, Ind10, Ind11, Ind12
Hausman Tests
To see if risk endogeneity is a problem, Hausman tests are conducted. Wage
regressions are estimated including the residuals from the risk estimations. If the
residual is significant, the null of exogeneity is rejected. Table 9 shows risk
endogeneity is a problem, with the residual variable significant in both estimations.
24
Table 9: Hausman Tests
Dependent Variables: Lnwpayl Lnwpayh
ALL MANUAL
Constant
4.7015***
(0.0462)
Runion
0.0998***
(0.0144)
Commspecific
0.0162
(0.0137)
Female
-0.2335***
(0.0147)
Fatal
3.7743***
(0.3509)
Fatal*Runion
-0.4613*
(0.2618)
Fatal*Commspecific
1.0162***
(0.2791)
Resid
-3.9312***
(0.3822)
Obs
Wald chi2
Log likelihood
5474
3280.91
-10933.072
MALE MANUAL
4.5624***
(0.0612)
0.0748***
(0.0181)
0.0298*
(0.0174)
3.3332***
(0.3607)
-0.2224
(0.3057)
0.8371***
(0.3258)
-3.6312***
(0.3910)
3754
1333.09
-7445.318
Other variables included in estimation: Educ1, Educ2, Educ3, Tenure2 Tenure3, Tenure4, Tenure5,
Overtime, Flexitime, Supervise, Permanent, Age, Age², Nemps, Nemps², Meritpay, Public
Controlling for Endogeneity
Following Garen, residuals from the risk estimations are included in the wage
regressions. Residuals are multiplied by the risk variable with this also included as an
explanatory variable. Key results are reported in Table 10.
Fatal remains positive and significant with the magnitude of the coefficients much
larger. This is reflected in the considerably higher VSL estimates. This is consistent
with the literature; OLS estimates that do not control for endogeneity appear to be
biased downwards. Sandy and Elliott (1996) observe this is consistent with the
argument that safety is a normal good, as “workers with high unobserved earnings
25
capacity are willing to pay for occupations with more safety” (p.300). The fact the
Fatal*Resid is significantly negative is also consistent with the literature, implying
“workers with unusually high risk have low values of unobserved earnings ability in
the presence of risk” (p.300). Other key conclusions remain once risk endogeneity is
controlled for, with health and safety committees having a positive effect and unions
having a negative impact upon the risk premium.
Table 10: Interval Regression Results Controlling for Endogeneity
Dependent Variables: Lnwpayl Lnwpayh
ALL MANUAL
Constant
4.7012***
(0.0463)
Runion
0.0950***
(0.0142)
Commspecific
0.0215
(0.0135)
Fatal
2.8489***
(0.3621)
Fatal*Runion
-0.4164*
(0.2523)
Fatal*Commspecific
0.8009***
(0.2692)
Resid
-3.2124***
(0.4152)
Fatal*Resid
-9.2013***
(2.0702)
MALE MANUAL
4.5606***
(0.0613)
0.0702***
(0.0178)
0.0384**
(0.0171)
2.2658***
(0.3762)
-0.0952
(0.2933)
0.5638*
(0.3170)
-2.8076***
(0.4421)
-10.2950***
(2.4620)
Obs
Wald chi2
Log likelihood
VSL (2004£)
VSL (2000 US$)
3754
1357.32
-7434.235
£40,920,438
$58,894,238
5474
3300.18
-10922.264
£46,587,093
$67,049,902
Other variables included in estimation: Educ1, Educ2, Educ3, Tenure2, Tenure3, Tenure4, Tenure5,
Overtime, Flexitime, Supervise, Permanent, Age, Age², Nemps, Nemps², Meritpay, Public
Instrument Tests
The instruments selected to proxy risk aversion are included in the wage estimation to
ensure they are appropriate. Most of the instruments are significant in the wage
26
estimations in both samples (with the exception of disability and some of the industry
dummies, and disability is never significant in the risk estimation) indicating they are
not appropriate instruments.
The problems encountered in the literature when attempting to control for risk
endogeneity are so large, that Bound et al. (1995) have suggested using weak
instruments results in coefficients that are biased towards the original OLS estimates.
To test if the instruments are weak, F tests are conducted. See Table 11.
Table 11: F Tests
Dependent variable
Married
Prob>F
Children
Prob>F
Disability
Prob>F
Pension
Prob>F
Car
Prob>F
Healthins
Prob>F
ALL MANUAL
0.11
(0.7446)
5.49
(0.0191)
0.03
(0.8552)
21.33
(0.0000)
37.18
(0.0000)
13.79
(0.0002)
MALE MANUAL
0.06
(0.8088)
5.91
(0.0151)
0.04
(0.8433)
22.01
(0.0000)
35.30
(0.0000)
11.07
(0.0009)
Staiger and Stock (1997) recommend that when testing the strength of an instrument,
the F statistic must take the value of at least 10. Similarly, Stock, Wright, and Yogo
(2002) found that an F statistic of 9 or above is needed for an appropriate instrument.
Table 11 reports married, children and disability are weak instruments in this case.
Pension, car and healthins however appear to be strongly correlated with a workers’
occupational risk. However, each of these three instruments were significantly
associated with pay.
27
Controlling for risk endogeneity may not be as essential to studies of compensating
wage differentials as some imply. Lalive et al. (2006) use Austrian longitudinal data
to estimate compensating wage differentials for risk of injury; the nature of their data
therefore controls for unobserved heterogeneity. They estimate a risk premium that is
roughly equal to the one obtained in a standard cross-sectional wage regression and
hence “find no evidence for a bias of the compensating differential obtained from a
standard cross-sectional hedonic wage function that can be attributed to unobserved
worker productivity” (p.4). Their results therefore suggest “the bias of the
compensating differential obtained from a standard cross-sectional hedonic wage
function that is due to unobserved productivity of workers or unobserved ability to
cope with risks is small” (p.19). This paper had access to very detailed data upon
work accidents and individual employment over time and so replication using other
international data may be difficult. It does however, suggest that risk premium
estimates obtained from cross sectional data with risk exogenous, may not be
misleading. Furthermore, although endogeneity may affect the magnitude of the risk
premium, the effect that institutions such as trade unions and health and safety
committees have upon the premium are unlikely to be altered by such potential bias.
6.
CONCLUSIONS
Manual workers do appear from these results to receive a wage premium for being
exposed to risk of death in British workplaces. Given that the only other UK cross
sectional studies use data from the 1980s, this is an important finding. VSL estimates
are calculated which enables the size of risk premiums to be compared with other
studies. For the sample of male manual workers, a VSL of approximately $13.2
million (in 2000 US$) is estimated, increasing to $58.9 million once we control for
28
endogeneity. Estimates vary widely between studies, as they depend upon many
factors such as what variables are included in the estimation, and whether endogenity
is controlled for. Viscusi and Aldy (2003) report estimates in US dollars relative to
the year 2000, results from key UK studies are reported below.
Table 12: UK VSL Estimates
Marin and Psacharopoulos (1982)
Siebert and Wei (1994)
Sandy and Elliott (1996)
Arabsheibani and Marin (2000)
Sandy et al. (2001)
Implicit VSL (million, 2000 US$)
$4.2
$19.6-$21.7
$5.2-$69.4
$19.9
$5.7-$74.1
Source: Viscusi and Aldy (2003)
Estimates calculated here therefore, are slightly higher than in the earlier literature but
do fit within the ranges. As with most studies using cross-sectional data, estimates
suffer from potential risk endogeneity. Although attempts to control for this bias are
adopted, here and in other papers, the problem of finding appropriate instruments to
proxy risk aversion limits the effectiveness of Garen’s method. Recent papers
however, have suggested the bias may not be as great as first thought.
Consistent with most of the British literature, trade unions are found to have a
negative effect upon the risk premium. This result is often explained by the suggestion
that unions are more concerned with increasing safety in the workplace rather than
bargaining for compensation for accidents. This explanation is supported by the
finding that firms that recognise a trade union have a smaller average risk rate. Most
significantly, the role of other health and safety institutions was considered, given that
their role may be increasing in importance in the workplace with the decline in the
presence of trade unions. Having a general committee that deals with a range of issues
29
or a safety representative with no committee had no impact upon the risk premium.
The presence of a committee that deals specifically with health and safety however,
had a positive impact upon the risk premium in the manual workers sample, with the
union effect remaining negative. This suggests that health and safety committees have
a significant role in influencing risk compensation and that they operate differently
and independently from trade unions in terms of health and safety. Health and safety
committees appear to have a positive impact upon the environment in which
bargaining takes place. In conclusion, these findings support Litwin’s (2000)
suggestion that the health and safety committee effect should be separated from the
union effect.
30
REFERENCES
ARABSHEIBANI, G R & MARIN, A (2000) ‘Stability of Estimates of the
Compensation for Danger’, Journal of Risk and Uncertainty, Vol. 20, No. 3, pp.247269
ASHENFELTER, O (2006) ‘Measuring the Value of Statistical Life: Problems and
Prospects’, The Economic Journal, Vol. 116, No.510, pp.10-23
BELL, D N., ELLIOTT, R & SANDY R (2004) ‘Unobserved Ability and
Compensating Differentials for Workplace Risks of Death’, Unpublished Paper
University of Stirling
BOUND, J., JAEGER, D A., & BAKER, R M (1995) ‘Problems with Instrumental
Variables Estimation When the Correlation Between the Instruments and the
Endogenous Explanatory Variables is Weak’, Journal of the American Statistical
Association, Vol. 90, No. 430, pp.443-450
DELEIRE, T & LEVY, H (2004) ‘Worker Sorting and the Risk of Death on the Job’,
Journal of Labor Economics, Vol. 22, No.4, pp.925-950
FAIRRIS, D (1992) ‘Compensating Wage Differentials in the Union and Nonunion
Sectors’, Industrial Relations, Vol. 28, No. 3, pp.205-221
FENN, P & ASHBY, S (2004) ‘Workplace Risk, Establishment Size and Union
Density’, British Journal of Industrial Relations, Vol. 42, No. 3, pp.461-480
GAREN, J (1988) ‘Compensating Wage Differentials and the Endogeneity of Job
Riskiness’, Review of Economics and Statistics, Vol. 70, No. 1, pp. 9-16.
HEALTH AND SAFETY EXECUTIVE (2006) ‘RIDDOR Explained’, Health and
Safety Executive
HECKMAN, J (1979) ‘Sample Selection Bias as a Specification Error’,
Econometrica, Vol. 47, No. 1, pp. 153-162
HWANG, H., REED, W R., HUBBARD, C (1992) ‘Compensating Wage
Differentials and Unobserved Productivity’, Journal of Political Economy, Vol. 100,
No.4, pp.835-858.
LALIVE, R., RUF, O., & ZWEIMULLER, J (2006) ‘Wages and Risks at the
Workplace: Evidence from Linked Firm-Worker Data’, Working Paper University of
Zurich
LITWIN, A S (2000) ‘Trade Unions and Industrial Injury in Great Britain’, Centre for
Economic Performance Working Paper Series, No.468
MACHIN, S (2000) ‘Union Decline in Britain’, British Journal of Industrial
Relations, Vol. 38, No. 4, pp.631-645
31
MARIN, A & PSACHAROPOULOS, G (1982) ‘The Reward for Risk in the Labour
Market: Evidence from the United Kingdom and a Reconciliation with Other Studies’,
Journal of Political Economy, Vo1. 90, No.4, pp. 827-853.
OFFICER, L H & WILLIAMSON, S H (2006) ‘Computing ‘Real Value’ Over Time
with a Conversion from British Pounds to US Dollars, 1830-2004’
MeasuringWorth.com August 2006
REILLY, B., PACI, P., & HOLL, P. (1995) ‘Unions, Safety Committees and
Workplace Injuries’, British Journal of Industrial Relations, Vol. 33, No.2, pp.275288
ROSEN, S (1986) “The Theory of Equalizing Differences” in Ashenfelter, O and
Layard, R (eds) Handbook of Labor Economics, Vol. 1, Elsevier Science Publishers,
pp.641-692.
SANDY, R & ELLIOTT, R F (1996) ‘Unions and Risk: Their Impact on the
Compensation for Fatal Risk’, Economica, Vol 63, pp. 291-309
SANDY, R & ELLIOTT, R F (2005) ‘Long-Term Illness and Wages: The Impact of
the Risk of Occupationally-Related Long-term Illness on Earnings’, The Journal of
Human Resources, Vol.40, No.3, pp.744-768
SANDY, R., ELLIOTT, R F., SIEBERT, W S., WEI, X., (2001) ‘Measurement Error
and the Effects of Unions on the Compensating Differentials for Fatal Workplace
Risks’ Journal of Risk and Uncertainty, Vol. 23, No. 1, pp.33-56
SIEBERT, W S & WEI, X (1994) ‘Compensating Wage Differentials for Workplace
Accidents: Evidence for Union and Non-union Workers in the UK’, Journal of Risk
and Uncertainty, Vol. 9, No. 1, pp.61-76.
SHOGREN, J F & STAMLAND, T (2002) ‘Skill and Value of Life’, Journal of
Political Economy, Vol. 110, No.3, pp. 1168-1173
VISCUSI, W K (1993) ‘The Value of Risks to Life and Health’, Journal of Economic
Literature, Vol. 31, No. 4, pp.1912-1946.
VISCUSI, W. KIP &ALDY, J E (2003) ‘The Value of a Statistical Life: A Critical
Review of Market Estimates Throughout the World’, Journal of Risk and
Uncertainty, Vol. 27, No.1, pp.5-76
32
APPENDIX 1: Explanatory Variables and Descriptive Statistics (Mean and
Standard Deviation)
VARIABLE
DEFINITION
MANUAL
Anincome
Annual Income
Wpayl
Lower value of weekly pay bracket.
Wpayh
Higher value of weekly pay bracket.
Educ1
Dummy variable equals one if highest
qualification is GCSE level.
Dummy variable equals one if highest
qualification A level.
Dummy variable equals one if respondent has a
degree.
Dummy variable equals one if respondent has no
academic qualifications (excluded in estimation).
Dummy variable equals one if respondent has
worked for the firm for less than one year
(excluded in estimation).
Dummy variable equals one if respondent has
worked for the firm for between 1 to less than 2
years.
Dummy variable equals one if respondent has
worked for the firm for between 2 to less than 6
years.
Dummy variable equals one if respondent has
worked for the firm for between 5 to less than 10
years.
Dummy variable equals one if respondent has
worked for the firm for 10 years or more.
Corresponds to the overtime or extra hours the
respondent usually works each week, paid or
unpaid.
Dummy variable equal to one if respondent has a
flexitime arrangement available to them if needed.
Dummy variable equal to one if respondent
supervises other employees.
Value between 0 and 4, with 0 indicating
employee very dissatisfied with job security, and 4
indicating they are very satisfied with their job
security.
Dummy variable equal to one if the respondent is
a permanent employee.
Equal to between 0-8, with 8 indicating the
employee is aged between 16-17 and a value of 8
indicating they are aged 65 or over.
Equal to the number of employees on the payroll
in the firm.
Dummy variable equal to one if some employees
within the firm receive merit pay.
Dummy variable equal to one if the respondent
works in the public sector.
Dummy variable equal to one if management
16352.66
(7082.163)
285.4982
(122.8716)
340.8521
(145.0228)
0.5070
(0.5000)
0.1199
(0.3249)
0.0914
(0.2882)
0.3254
(0.4686)
0.1407
(0.3477)
MALE
MANUAL
18060.04
(6982.503)
315.4544
(120.3039)
376.0072
(143.3464)
0.5038
(0.5000)
0.1145
(0.3185)
0.0897
(0.2858)
0.3296
(0.4701)
0.1322
(0.3388)
0.1095
(0.3123)
0.0996
(0.2995)
0.2344
(0.4237)
0.2265
(0.4186)
0.2022
(0.4016)
0.2030
(0.4023)
0.3115
(0.4631)
4.4294
(7.0093)
0.3375
(0.4729)
4.9074
(7.2196)
0.2566
(0.4368)
0.2482
(0.4320)
2.5052
(1.0434)
0.2343
(0.4236)
0.2492
(0.4326)
2.4374
(1.0506)
0.9448
(0.2284)
4.6405
(1.4130)
0.9542
(0.2090)
4.7037
(1.3837)
360.8466
(730.4300)
0.2283
(0.4198)
0.2129
(0.4089)
0.5256
365.4467
(683.7581)
0.2417
(0.4281)
0.1663
(0.3724)
0.5609
Educ2
Educ3
Educ4
Tenure1
Tenure2
Tenure3
Tenure4
Tenure5
Overtime
Flexitime
Supervise
Security
Permanent
Age
Nemps
Meritpay
Public
Unioncov
33
Runion
Commspecific
Commgen
Emprep
Nohsconsult
report recognising a trade union for negotiating
pay and conditions
Dummy variable equal to one of the respondent
works for a firm where there is a trade union.
Dummy variable equal to one if the workplace has
a safety representative
Dummy variable equal to one if the workplace has
a consultative committee that deals with a range of
issues including health and safety.
Dummy variable equal to one if the workplace has
a specific health and safety committee.
Dummy variable equal to one if management
deals with health and safety matters without any
form of consultation.
(0.4994)
(0.4963)
0.5575
(0.4967)
0.4186
(0.4934)
0.0636
(0.2441)
0.5758
(0.4943)
0.4462
(0.4972)
0.0705
(0.2561)
0.2443
(0.4299)
0.2735
(0.4458)
0.2475
(0.4732)
0.2358
(0.4246)
34
APPENDIX 2: Key Interval Regression Results (Fatal, Commspecific,
Commgeneral)
Dependent Variables: Lnwpayl Lnwpayh
ALL MANUAL
Constant
4.8419***
(0.0443)
Runion
0.0925***
(0.0145)
Commspecific
0.0274*
(0.0142)
Commgeneral
0.0334
(0.0252)
Female
-0.3279***
(0.0252)
Fatal
0.5452***
(0.1980)
Fatal*Runion
-0.5692**
(0.2700)
Fatal*Commspecific
0.5939**
(0.2844)
Fatal*Commgeneral
0.0420
(0.5040)
Obs
Wald chi2
Log pseudo likelihood
VSL (2004 £)
VSL (2000 US$)
5474
3111.46
-10984.538
£13,628,307
$19,614,374
MALE MANUAL
4.7042***
(0.0585)
0.0581***
(0.0181)
0.0430**
(0.0180)
0.0618**
(0.0304)
0.4572**
(0.2195)
-0.2363
(0.3077)
0.3604
(0.3255)
-0.2213
(0.5443)
3897
1269.82
-7780.617
£14,150,041
$20,365,273
Other variables included in estimation: Educ1, Educ2, Educ3, Tenure2 Tenure3, Tenure4, Tenure5,
Overtime, Flexitime, Supervise, Permanent, Age, Age², Nemps, Nemps², Meritpay, Public
35
APPENDIX 3: Risk Instrumental Variables and Descriptive Statistics (Mean and
Standard Deviation)
VARIABLE
DEFINITION
Married
Equal to 1 if the worker is married or
living with a partner.
Equal to 1 if the worker has dependent
children (aged 0-18)
Equal to 1 if the worker describes
themselves as having a long-term illness,
health problem, or disability.
Equal to 1 if manager reports workers in
the largest occupational group are entitled
to an employee pension scheme.
Equal to 1 if manager reports workers in
the largest occupational group are entitled
to a company car or car allowance.
Equal to 1 if manager reports workers in
the largest occupational group are entitled
to private health insurance.
Equal to 1 if workplace in manufacturing
industry.
Equal to 1 if workplace in electricity
industry.
Equal to 1 if workplace in construction
industry.
Equal to 1 if workplace in wholesale
industry (excluded).
Equal to 1 if workplace in hotel industry.
Children
Disability
Pension
Car
Healthins
Ind1
Ind2
Ind3
Ind4
Ind5
Ind6
Ind8
Equal to 1 if workplace in transport
industry.
Equal to 1 if workplace in financial
industry.
Equal to 1 if workplace in other industry.
Ind9
Equal to 1 if workplace in public industry.
Ind10
Equal to 1 if workplace in education
industry.
Equal to 1 if workplace in health industry.
Ind7
Ind11
Ind12
Equal to 1 if workplace in other
community industry.
ALL
MANUAL
0.6683
(0.4709)
0.3737
(0.4838)
0.1385
(0.3455)
MALE
MANUAL
0.6903
(0.4624)
0.4200
(0.4936)
0.1413
(0.3484)
0.7776
(0.4159)
0.7912
(0.4065)
0.1204
(0.3255)
0.1198
(0.3248)
0.1398
(0.3255)
0.15091
(0.3580)
0.3142
(0.4642)
0.0201
(0.1403)
0.0692
(0.2538)
0.0713
(0.2574)
0.0367
(0.1881)
0.1344
(0.3411)
0.0036
(0.0598)
0.0665
(0.2492)
0.0181
(0.1333)
0.0525
(0.2231)
0.1522
(0.3592)
0.0613
(0.2400)
0.3678
(0.4823)
0.0265
(0.1608)
0.0948
(0.2930)
0.0872
(0.2822)
0.0255
(0.1578)
0.1663
(0.3724)
0.0033
(0.0572)
0.0756
(0.2644)
0.0187
(0.1355)
0.0245
(0.1547)
0.0485
(0.2149)
0.0612
(0.2397)
36